The climate models that were discarded for predicting too much warming were right, and the ones the world's climate projections rest that were the basis of the Paris Agreements targets are all wrong, say scientists, backed up by the data.

The Paris Agreement cap of 1.5°C increase in temperatures goal is based on the calculations of around 20 climate model predictions coordinated by the United Nations’ Intergovernmental Panel on Climate Change (IPCC) that has formed the baseline for the last decade. But as the Climate Crisis gets underway in earnest over the last four years of what is now an annual disaster season, it is already clear that global warming is running faster than any of the models predicted. The climate model calculations are badly off.

The Climate Crisis is accelerating. The IPCC says that the Paris Agreement goal of keeping temperature increases to less than 1.5°C-2°C above the pre-industrial benchmark has already been missed and temperature increase are on course to reach a catastrophic 2.7C-3.1C by 2050. At that point extreme temperature events will become routine and large parts of the world will become uninhabitable.

A new study by a team at ETH Zurich, “Recent Temperature and Energy Imbalance Trends Point to Higher Estimates of Future Warming”, published in the journal Earth's Future, rebuilds the way climate models are graded against reality and finds that the best-performing ones project substantially more warming than the current consensus.

"For a given emission trajectory, we expect warming to be 25% higher," said Gergana Gyuleva, who led the study.

Every climate target in circulation is a temperature and every temperature comes out of these models. The Climate Action Tracker currently puts the world on course for 2.6°C above pre-industrial levels by 2100 if countries do what they have already promised – which most of them are not doing. On Gyuleva's team's numbers the same policies deliver 3.25°C - which is the difference between a bad century and an unrecognisable one, produced not by any change in emissions but by a change in which models are trusted. As IntelliNews reported, if the permafrost melts in Russia and the north completely disappears, as it will at a 3°C increase above the pre-industrial baseline, that leads to catastrophic and unpredictable consequences.

How the models got graded

Climate models disagree sharply about how much the planet heats for a given amount of carbon dioxide. Some put the short-term response to a doubling of CO₂ at around 1.6°C; others at close to 3°C. "The range is big, and it really matters," said Gyuleva. "So, the point is to find out which ones are right."

The standard test has been to run the models over the recent past. Feed in the last century's greenhouse gas emissions, compare the warming the model produces against the warming that actually happened, and keep the ones that match.

For the IPPC’s last report, a lot of models simulated more warming than had been observed, so they were discounted. What survived gave a short-term response to doubled CO₂ of between 1.2°C and 2.4°C, with a best estimate of 1.8°C.

The problem with grading models against the observed record is that the observed record contains weather as well as climate. During a La Niña, cooler water from the deep ocean spreads across the surface and global temperatures dip; the reverse happens in an El Niño. None of that says anything about how sensitive the climate is to CO₂ .

Gyuleva's team filtered the historical record to strip that variability out, and found it had been pushing in one direction. Natural variability suppressed measured warming between 1981 and 2014 - a window that includes the so-called global warming hiatus of the 2000s, and the exact window the earlier model-grading papers used.

"It's really exactly this period 1981 to 2014, which was used in the previous papers, where you see the strongest bias down," said Gyuleva. "It was really bad luck."

Correcting for that alone raises the estimate of how much warming a doubling of CO₂ produces. The models were not too hot. The three decades they were being judged against were unusually cool.

Measuring the heat, not the thermometer

The team then added a second test that does not use surface temperature at all. Satellites have been measuring how much short-wave radiation from the sun enters the atmosphere – the earth’s energy imbalance (EEI) - and how much long-wave radiation leaves it. The planet's energy imbalance is the most fundamental measure of global warming there is, and one that does not care where the ocean has moved its heat this decade.

Because the satellite measurements only start in 2001, this kind of assessment has become possible only recently. How well a model reproduces the observed trends in incoming and outgoing radiation is a harder test than how well it reproduces surface temperature.

The models that best match those radiation trends project much more warming. Taking the models that perform well on both tests, the team puts the short-term response to a doubling of CO₂ at between 1.9°C and 2.6°C, with a best estimate of 2.25°C - against the IPCC's 1.8°C.

"Our results need to be confirmed by more evidence from different sources before we can be fully confident," Gyuleva said.

What other scientists make of it

The reception has been warm without being unanimous. "This seems to be a sound approach," said Trude Storelvmo at the University of Oslo. "In my opinion, the filtering out of natural variability is a valuable step forward," said Drew Shindell at Duke University in North Carolina - though he is not yet convinced the radiation trends make a good yardstick for model performance.

Steven Sherwood at the University of New South Wales in Sydney called it "a plausible conclusion and a reasonable approach", while noting the ceiling on any such exercise.

"To do a better job of this we really need better climate models," Sherwood said. "On the other hand, the steep temperature rises of the last few years are going to lead to higher estimated ranges regardless of method."

That last observation is the one that should worry policymakers most. Whichever method wins the methodological argument, the direction of revision is the same.